Target detection technology widely applied in many fields has always been an important research subject in image processing and computer vision. As a front end processing link, the target detection is one key technology and difficult problem. Only the target in a scene is detected timely, a series of processing work in the following such as target tracking and identifying can be implemented smoothly.
Since target detection technology is very susceptible to the external light especially for sunlight, when a target detection apparatus is used outside, the measurement accuracy will be seriously affected with the detected target in the presence of relatively strong background light, especially under the perpendicular incidence of sunlight or strong light. In addition, positional deviation caused by the shaking and movement of the target to be detected would bring about great impact on the detection accuracy during detection.
How to prevent the relatively strong light from interfering the target detection as well as how to reduce the deviation for the target detection during movement have always been a bottleneck problem and technical difficulty to be solved which puzzle and restrict the practicality of target detection system. Right now, great attention has been paid by domestic and foreign experts to it, and thorough and broad studies have been carried out on such subject. In the present invention, solutions are put forward mainly aiming at such two challenges.
In the prior art, there are three kinds of technical solutions to prevent the sunlight interference and the influence of target movement in the target detection technology, all of which are relevant to the present invention.
In a first method, an additional filter clogging plate or a sunlight occlusion device is utilized. A China patent with application number of No. 200820052299 provides a light curtain that can be used under the sunlight outside. The light curtain comprises a microcontroller which is connected with an IR receiver and an IR transmitter respectively. The body of the IR receiver is provided with a dimmer on its front side so as to filter out visible light as well as reserve IR and far IR signals. However, the above mentioned method cannot solve problems by the root and eliminate the influence of moving objects on the target detection. Besides, the filtrating would cause a decrease of the sensitivity as well. As such, the system has to improve the transmitting power to offset the decrease of sensitivity induced by filtering. In addition, the additional filter clogging plate or the occlusion device to the system will lead to an increase in cost and inconvenience in use.
In a second method, it is a target detection method based on background subtraction method. The basic idea for the background subtraction method is as follows: an average of one or several images in the background is selected to be a background image firstly; then take difference between a current frame and the background, and the obtained difference image is processed appropriately; if the obtained pixel number is larger than a certain threshold value, then a target is determined to be in the background. In this way, the target can be extracted precisely. However, such method is susceptible to many factors including illumination in the background and the movement of background object, as such the produced shadow of moving object will considerably influence the detection accuracy.
The third method is a three-frame difference method. A China patent with application number of No. 200610021067 discloses a moving target detection method in infrared image sequence in complex background, which belongs to the technical field of digital image processing. At first, calculation is performed to obtain difference images of a previous frame and a current frame as well as a current frame and a next frame in the infrared image sequence respectively. After that, a respective change detection template is extracted from the two difference images. Subsequently, a union operation is carried out on the two change detection templates, and their intersection is taken as a moving target template. Finally, a connected component labeling is implemented to the moving target template to obtain the moving target. Wherein the extract of the change detection template specifically comprises: extracting an original change detection template by estimating noise parameters for the difference images in a manner of iteration from both ends of a column diagram to its centre; employing a morphologic processing mode including performing an and operation on small structural units after their erosion and dilation and performing a closing operation on large structural units; wherein the noise parameters for the difference images comprise mean values and variances. Although the three-frame difference method can eliminate the target deviation produced during movement effectively and achieve relatively precise moving target detection, it has not resolved the impact of strong light interference on the target detection. Besides, the algorithm is complex in implementation, in which it is needed to carry out a great deal of morphologic processing to the image to reduce noise in the last period.
During the detection for moving target, the sweep time also has great influence on the detection result. In the detection methods as described above and any other detection methods, it usually takes about 15 ms to scan one frame image. However, the moving object may have experienced large replacement in such time period. Thus, the image acquired in this way and the post image processing will have great deviation.